2018-01-23 05:35:45

by Dave Jones

[permalink] [raw]
Subject: problematic rc9 futex changes.

c1e2f0eaf015fb: "futex: Avoid violating the 10th rule of futex" seems to
make up a few new rules to violate.

Coverity picked up these two problems in the same code:


First it or's a value with stack garbage.

_______________________________________________________________________________________________________
*** CID 1427826: Uninitialized variables (UNINIT)
/kernel/futex.c: 2316 in fixup_pi_state_owner()
2310
2311 raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
2312
2313 oldowner = pi_state->owner;
2314 /* Owner died? */
2315 if (!pi_state->owner)
>>> CID 1427826: Uninitialized variables (UNINIT)
>>> Using uninitialized value "newtid".
2316 newtid |= FUTEX_OWNER_DIED;
2317
2318 /*
2319 * We are here because either:
2320 *
2321 * - we stole the lock and pi_state->owner needs updating to reflect

Then it notices that value is never read from before it's written
anyway.

*** CID 1427824: Code maintainability issues (UNUSED_VALUE)
/kernel/futex.c: 2316 in fixup_pi_state_owner()
2310
2311 raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
2312
2313 oldowner = pi_state->owner;
2314 /* Owner died? */
2315 if (!pi_state->owner)
>>> CID 1427824: Code maintainability issues (UNUSED_VALUE)
>>> Assigning value from "newtid | 0x40000000U" to "newtid" here, but that stored value is overwritten before it can be used.
2316 newtid |= FUTEX_OWNER_DIED;
2317
2318 /*
2319 * We are here because either:
2320 *
2321 * - we stole the lock and pi_state->owner needs updating to reflect


(The next reference of newtid being..

2369 newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;


Dave


2018-01-23 08:34:36

by Peter Zijlstra

[permalink] [raw]
Subject: Re: problematic rc9 futex changes.

On Tue, Jan 23, 2018 at 12:34:46AM -0500, Dave Jones wrote:
> c1e2f0eaf015fb: "futex: Avoid violating the 10th rule of futex" seems to
> make up a few new rules to violate.
>
> Coverity picked up these two problems in the same code:
>

Yeah, Geert also spotted it:

https://lkml.kernel.org/r/[email protected]

I've been running the robustpi tests from glibc but have so far failed
to actually trigger the bug. I think I'll just write up a Changelog and
post the fix from the above link.